Skip to content
Inexra AI assistant

Drop in 10 posts. Get back a scheduled month.

Inexra AI is an actual agent — not a chat box. It calls workspace tools, drafts captions in each client's brand voice, writes the first-comment hashtags, and proposes a full schedule. You approve before anything ships.

Free Creator plan · No credit card · Approval gate on every action

One brief, one batch

What happens when you say "schedule these 10 reels."

Inexra AI assistant

From brief to scheduled — in one chat.

Human approval before anything ships
  1. 01

    Drop in 10 posts

    You

    “Schedule these 10 reels across the next two weeks, Tue/Thu mornings, in the client’s voice.” Upload the media. That’s the brief.

  2. 02

    Captions written in brand voice

    Assistant

    getBrandVoice → findContentTopics → drafts a caption per post tuned to the client’s tone, audience, and last 30 days of top performers.

  3. 03

    First comment + hashtags

    Assistant

    Adds a stacked-hashtag first comment (no caption clutter) and a clean CTA — sized per post so reach hashtags don’t cap the wrong reel.

  4. 04

    Slots picked, schedule proposed

    Assistant

    proposePostSchedule queues every post into the calendar at the cadence you asked for — no double-booked slots, no past timestamps.

  5. 05

    Approve once, all 10 ship

    You

    Proposal-token gate: review every caption, hashtag, and slot in one screen. Approve to publish, edit before you do, or reject — nothing leaves the workspace until you say so.

Tools the assistant uses

searchMessagesgetInboxSummarygetPipelineStatusgetBrandVoicefindContentTopicsproposePostScheduleproposeCancelScheduledPostsgetRecentScheduledPostsremember

Guardrails

Per-turn quota, plan AI caps, action logging — including Anthropic usage events. Nothing runs without a signed proposal token you approve.

What the assistant actually does

Four jobs, one approval screen.

Captions written in brand voice

getBrandVoice reads each client's tone profile, findContentTopics surfaces what their audience has actually engaged with — then the assistant drafts a caption per post tuned to both.

First-comment hashtags, sized per post

Hashtag stacks land in the first comment (no caption clutter) and are scaled per post — broad reach where it helps, niche where it converts. Plus a clean CTA if the post asks for one.

Calendar slots proposed for the whole batch

proposePostSchedule queues every post at the cadence you asked for — Tue/Thu mornings, alternating reels and carousels, whatever the brief said. No double-booked slots. No past timestamps.

Proposal-token approval gate

Nothing reaches the platform until you click approve. Every proposal — captions, hashtags, slots — is reviewable, editable, rejectable in one screen, all 10 posts at once.

The toolbelt

Real tools. Real workspace data. Real approvals.

The assistant isn't a stateless chat — it invokes named tools against the live workspace. Every call is logged, every action that touches the platform is gated.

searchMessages

Find every DM matching a brand, topic, or thread — across a workspace.

getInboxSummary

Today's queue at a glance: brand inquiries, complaints, sentiment, what needs a reply.

getPipelineStatus

Live deal stages, stuck deals, what closes this month.

getBrandVoice

Reads each client's tuned voice profile before drafting anything.

findContentTopics

Surfaces themes the client's audience has actually engaged with.

proposePostSchedule

Stages up to 20 posts at once with captions, media, first-comment, and slots.

proposeCancelScheduledPosts

Pulls a scheduled post (or batch) back out — also gated by approval.

getRecentScheduledPosts

Lists what's already queued so the assistant doesn't double-book.

remember

Persists client preferences and decisions to long-term workspace memory.

Why it's safe to actually use

Guardrails that ship with the assistant.

Agents that touch real client accounts need real controls. These are on by default — not feature-flagged, not an enterprise upgrade.

  • Proposal-token signing — actions can't execute without an approved token
  • Per-turn quota stops runaway loops
  • Plan-based AI question + reply caps enforced at the API boundary
  • Anthropic usage events logged per workspace for billing transparency
  • Action log records every tool call, input, and outcome
  • Workspace isolation — the assistant only ever sees one client's data at a time

Things you can actually say to it

Prompts that turn into proposals.

You say

“Schedule these 10 reels Tue/Thu mornings for two weeks in the client's voice — first-comment hashtags, CTA on every reel.”

Assistant proposes

Proposes 10 ScheduledPost rows with captions, first-comment hashtag stacks, and slot times. One approval screen, all 10.

You say

“Who from the brand inbox needs a reply today?”

Assistant proposes

searchMessages + getInboxSummary returns the queue scoped to the active workspace — sorted by priority, with urgency tags.

You say

“Cancel the carousel scheduled for Friday and reschedule the rest tighter.”

Assistant proposes

proposeCancelScheduledPosts stages a removal + proposePostSchedule restages the rest. You approve both at once.

The assistant works inside the agency stack

Plug it into isolated client workspaces.

Each client gets their own workspace — brand voice, history, tools. The assistant only ever sees one client at a time, with role-based team permissions and agency-wide rollups.

See it propose a real month

Bring 10 posts. Leave with a scheduled month.

On the demo we'll plug in one of your real client accounts, hand the assistant a batch of posts, and walk through the approval screen together.